A Comparative Analysis of Time Series Prediction Models for Energy Consumption in Primary Food Processing

Energy consumption forecasting is a key challenge for primary food processing companies aiming to optimize operations, reduce costs, and improve sustainability. This study compares the performance of different time series forecasting models applied to energy consumption data from a frozen vegetable processing plant in Spain. The models analyzed include statistical approaches such as Seasonal AutoRegressive Integrated Moving Average (SARIMA), deep learning methods including Long Short-Term Memory (LSTM) and transformer-based models. The dataset used combines historical energy consumption, production data, and weather conditions, forming a multivariate time series to improve predictive accuracy. The model underwent through several experiments to achieve satisfactory results, this included univariate experiments, univariate experiments with complex models and multivariate with a monthly and a weekly granularity. Results demonstrate that while traditional models provide a reliable baseline, machine learning and deep learning methods capture more complex patterns, offering superior performance under varying operational and environmental conditions. This comparative analysis highlights the importance of selecting appropriate forecasting techniques for effective energy management in the food processing industry, contributing to both economic and environmental sustainability.